3D CT to 2D low dose single-plane fluoroscopy registration algorithm for in-vivo knee motion analysis

A limitation to accurate automatic tracking of knee motion is the noise and blurring present in low dose X-ray fluoroscopy images. For more accurate tracking, this noise should be reduced while preserving anatomical structures such as bone. Noise in low dose X-ray images is generated from different sources, however quantum noise is by far the most dominant. In this paper we present an accurate multi-modal image registration algorithm which successfully registers 3D CT to 2D single plane low dose noisy and blurred fluoroscopy images that are captured for healthy knees. The proposed algorithm uses a new registration framework including a filtering method to reduce the noise and blurring effect in fluoroscopy images. Our experimental results show that the extra pre-filtering step included in the proposed approach maintains higher accuracy and repeatability for in vivo knee joint motion analysis.